Personalized and Course-Specific Performance Forecasting Model in Cycling

Abstract

In elite cycling, due to the influence of personal and terrain factors, performance forecasting needs to be personalized and course-specific. We design and empirically assess a variety of personalized and course-specific performance forecasting models based on random forest, feed forward neural network (FFNN), recurrent neural networks (RNNs), and long short term memory (LSTM). The mean square error (MSE) is selected as the metric for model comparison. Experiments shows that the LSTM models have the lowest MSE on both the heart rate forecasting and speed forecasting on the test dataset.